Possible fix for world-to-pixel when out of bounding box

[bmorris3]

This PR attempts to address #496.

GWCS implements an in_image method that returns a False for out-of-bounds world coordinates by converting to pixel coords with invert and then comparing to the bounding box.

This PR simply adds a line to world_to_pixel and world_to_pixel_values which sets pixel results outside of the bounding box to nan. Rather than calling in_image, I moved the logic for pixel coordinate comparisons to the bounding box into a separate function, and call that function in world_to_pixel* transformations, as well as within in_image. The pixel/bounding box comparison logic had to be moved to a separate method because in_image calls invert – so calling in_image to from within invert to mask the result would lead to recursion.

Right now there are two test failures where I would have to change the expected output in the test. I've left these two cases in so you can see what changes. Let me know if you agree that I can change those expected answers to nan.

cc @astrofrog @nden

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 87.37%. Comparing base (eb9d316) to head (9068f7a).
Report is 6 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #497      +/-   ##
==========================================
+ Coverage   87.28%   87.37%   +0.09%     
==========================================
  Files          22       22              
  Lines        3821     3834      +13     
==========================================
+ Hits         3335     3350      +15     
+ Misses        486      484       -2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.